Simple way to control IK solution configurations

[petercorke]

One of the problems with numerical IK is that you can't control the solution of the arm, it will be randomly elbow up or elbow down depending on how the RNG dice roll.

I wonder (pure speculation) if we could get some control of the configuration by controlling the random configuration generator so that configuration vectors are not uniformly distributed over the joint angle range, but some subset. For example, if we want an elbow up solution, we should only allow random seeds where the shoulder joint has the upper arm above the horizontal plane, not below. How that's expressed in terms of joint configuration vector depends on arm structure and zero-angle conventions...

If this worked, probably easiest to test on the Python version first, then there's the challenge of how to express this? We could:

• pass in a custom random configuration generator
• pass in a function that tests a random configuration generated by the IK solver and either allows it or requests another one
• passes in some string like "q[2]>0; -pi<q[3]<0" which is parsed by the IK solver.

There would be some expense in having the C++ code making Python callbacks.